Key Takeaways
- Strategic alliance: Nscale and humanoid‑robot maker Figure have locked in a multi‑year partnership that makes Nscale Figure’s preferred compute provider and a shareholder.
- GPU scale: The deal targets the deployment of up to 100,000 NVIDIA Vera Rubin GPUs, beginning in the second half of 2027 at a dedicated data center in Barstow, Texas.
- Financial commitment: Figure will initially spend US $3.5 billion on compute resources, with a roadmap to exceed US $6 billion as the robot fleet expands.
- Vertical AI integration: Nscale will deliver the power, networking, and orchestration stack, embedding NVIDIA’s AI‑optimized hardware and software to accelerate Figure’s Helix humanoid platform from training through real‑world operation.
- Supply‑chain synergy: Both companies will explore joint sourcing of components for future humanoid production, potentially lowering BOM costs and lead times.
Overview of the Partnership
Why Nscale and Figure Joined Forces
Figure, known for its Helix series of bipedal robots, requires massive compute horsepower to run perception, planning, and control algorithms in real time. Nscale specializes in large‑scale AI infrastructure and has a proven track record delivering high‑density GPU farms for autonomous‑vehicle and cloud‑AI customers. By combining Figure’s robotics expertise with Nscale’s data‑center capabilities, the two firms aim to create a vertically integrated AI pipeline that minimizes latency and maximizes throughput for physical AI workloads.
Deployment Timeline & Geography
- Location: A purpose‑built facility near Barstow, Texas, chosen for its low‑cost power grid and proximity to major West‑coast data routes.
- Phase 1 (H2 2027): Installation of the first 30,000 NVIDIA Vera Rubin GPUs, delivering roughly 1.5 PFLOPS of mixed‑precision AI performance.
- Phase 2 (2028‑2029): Ramp‑up to the full 100,000‑GPU capacity, pushing total compute beyond 5 PFLOPS and supporting up to 10,000 concurrent Helix units in the field.
Technical Foundations
NVIDIA Vera Rubin Platform
| Feature | Specification | Relevance to Humanoid AI |
|---|---|---|
| GPU Model | H100‑NVL (customized for Vera Rubin) | 2× higher tensor‑core density than standard H100, ideal for multi‑modal sensor fusion |
| Memory | 80 GB HBM3 per GPU | Enables large‑scale transformer models (>2 B parameters) to run locally on the robot |
| Interconnect | NVLink 4.0 + NVSwitch | Sub‑microsecond communication across GPU clusters, critical for real‑time control loops |
| Power Envelope | 300 W per GPU (optimized cooling) | Supports dense rack designs while keeping PUE < 1.15 |
Nscale’s Orchestration Layer
- Auto‑Scaling Scheduler: Dynamically allocates GPU slices based on robot fleet demand, reducing idle capacity to < 5 %.
- Edge‑to‑Cloud Sync: Secure, low‑latency pipelines that push model updates from the Barstow hub to on‑board robot modules within 200 ms.
- Security Stack: End‑to‑end encryption and hardware root of trust, meeting ISO 27001 and IEC 62443 standards for industrial robotics.
Financial & Market Implications
- Initial spend: $3.5 B earmarked for hardware procurement, rack construction, and power contracts.
- Growth ceiling: Projections indicate a total outlay of >$6 B by 2030, driven by scaling the Helix line to commercial sectors such as logistics, construction, and elder‑care.
- Shareholder alignment: Nscale’s equity stake in Figure aligns incentives, encouraging joint R&D on next‑gen perception chips and lightweight chassis designs.
- Competitive edge: The combined compute density (≈ 5 PFLOPS) surpasses most public cloud AI offerings for robotics, potentially lowering per‑robot inference cost from $0.12 to $0.07 per inference hour.
Comparison with Alternative Compute Providers
| Provider | GPU Count (Planned) | Avg. GPU Power (W) | Estimated PFLOPS | Typical Latency (ms) | Pricing (per GPU‑hour) |
|---|---|---|---|---|---|
| Nscale (Figure) | 100,000 | 300 | >5 | 0.2 (edge‑cloud) | $0.04 |
| AWS Graviton‑AI | 45,000 | 350 | 2.3 | 0.8 | $0.07 |
| Google Vertex AI | 38,000 | 340 | 1.9 | 0.6 | $0.06 |
| Azure AI Supercomputer | 50,000 | 320 | 2.5 | 0.7 | $0.065 |
Source: Company press releases, public cloud pricing sheets (2024‑2025).
Future Outlook
Both parties have signaled intent to co‑develop a “humanoid‑centric” supply chain, potentially leveraging Nscale’s bulk‑ordering clout to secure lower‑cost silicon, power modules, and sensor arrays. If successful, the partnership could set a new benchmark for “Physical AI” – AI that directly interacts with the world – by delivering compute at a scale traditionally reserved for data‑center‑only workloads.
Bottom Line
The Nscale‑Figure alliance couples a massive, NVIDIA‑driven GPU infrastructure with a leading humanoid robot platform, backed by a multi‑billion‑dollar commitment. By deploying up to 100,000 Vera Rubin GPUs in Texas, the partnership promises sub‑200 ms end‑to‑end latency, unprecedented compute density, and a clear path to scaling the Helix robot fleet across multiple industries. For investors and industry watchers, the deal marks a decisive move toward tightly integrated AI hardware and robotics, potentially reshaping the economics of physical AI deployment.